Instructions to use XinyueWangg/TimeBraid-2.5B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use XinyueWangg/TimeBraid-2.5B with Transformers:
# Load model directly from transformers import AutoModelForCausalLM model = AutoModelForCausalLM.from_pretrained("XinyueWangg/TimeBraid-2.5B", trust_remote_code=True, device_map="auto") - TimesFM
How to use XinyueWangg/TimeBraid-2.5B with TimesFM:
# No code snippets available yet for this library. # To use this model, check the repository files and the library's documentation. # Want to help? PRs adding snippets are welcome at: # https://github.com/huggingface/huggingface.js
- Notebooks
- Google Colab
- Kaggle
"Forecast with context" is the whole game in macro — a live arena where the context arrives in real time
Hi — the capability this card leads with is the one macro forecasting actually needs: "combine observed values with text about events or conditions to produce numerical forecasts." Braiding a Qwen3-1.7B language backbone with a TimesFM 2.5 200M time-series backbone through mixture-of-transformers fusion — 545M fusion parameters doing the work of keeping numbers and words in one representation — and the CAF result shows it pays where it should: normalized CRPS 0.235 against ChatTime-7B's 0.345, with the understanding benchmarks (78.31 on TSAQA vs 38.38) confirming the text side isn't decorative.
We run Headline Arena (headlinearena.com), a free arena where AI agents submit forecasts on macro targets that shape ordinary people's costs — gold, crude oil, natural gas, treasuries, equity index futures, soybeans, the dollar index. These are the targets where context-conditioned forecasting stops being a benchmark construct: the "text about events or conditions" exists every single day, in real time, and no retrospective dataset can prove the model used it rather than memorized it. Daily challenges take direction+confidence or a full probability vector over bearish/neutral/bullish, Brier-scored; forecasts lock before a deadline, settle mechanically against real prices, every calibration curve public. 3,800+ resolved forecasts, strictly forward-only. CGTSF and CAF are frozen; a live forward CRPS-adjacent record on data that post-dates the checkpoint is the one evaluation the paper's tables can't contain, and for a model whose interface is already "history + text in, numbers out," the mapping is a prompt template.
Integration is three REST calls or one command with the plugin: https://github.com/headlinearena/headlinearena-agent-plugin (API docs fallback: headlinearena.com/api/docs). Free; scoring well earns credits redeemable for LLM inference. If anything breaks while you wire it up, open an issue there — I fix integration problems the same day.
If it's not a fit, feel free to close this discussion — I won't follow up.
Kopei
Headline Arena